Want to bet on the future of artificial intelligence? Nvidia is a solid choice. It designs chips and software that run AI, and is popular on the stock market, rising 144% this year alone.
But some stock traders have found a less obvious backdoor option: utility stocks, particularly those of companies that own nuclear power plants.
Two such companies, Constellation Energy and Vistra, were the top performers on the S&P 500 index in September with returns of more than 30%. Shares of Constellation Energy, the largest U.S. nuclear power plant operator, soared on Sept. 20 as it supplies Microsoft’s fast-growing AI data center with energy from the Three Mile Island nuclear power plant near Harrisburg, Pennsylvania. The stock price soared after the deal was signed.
If you followed the news closely in 1979 or were familiar with nuclear history, you would know that Three Mile Island was the site of the worst nuclear disaster in U.S. history. Now, because of AI’s energy needs, the utility is reopening the undamaged portion of Three Mile Island that was saved five years ago and renaming it the less exciting Crane Clean Energy Center. It’s planned.
Vistra also said it is in talks with multiple AI operators. And the Constellation deal with Microsoft followed a March AI and nuclear power deal involving Talen Energy and Amazon. Mr. Talen agreed to sell large amounts of electricity to Amazon from the Susquehanna Steam Power Plant, a nuclear power plant near Berwick, Pennsylvania. The power plant holds 90% ownership of the power plant.
Consider the annual stock return, including dividends.
The S&P 500 stock index is 22%.
Constellation Energy, 121 percent;
Vistra, 208 percent;
Talen Energy, 185 percent;
Other utilities are also benefiting from AI, but generally not as spectacularly as these companies.
Restarting more idle nuclear power plants, such as Michigan’s Palisades and Iowa’s Duane Arnold, is being considered, but this is a difficult and expensive process that would require huge investments and regulatory approvals. It’s a process. Nuclear power plants that operate for other purposes and provide electricity may be transitioned to more profitable operations using AI data centers in the future.
Dominion Energy CEO Robert Blue said on a conference call last month that the company “is not interested in the idea” of doing such a deal at the Millstone power plant, which began operations in Waterford, Conn., in 1970. I’m certainly positive about that.”
Nuclear power plants are not the only plants whose lives could be extended. Old coal-fired power stations slated for decommissioning are being given new life. The growing demand for electricity from AI data centers is outpacing energy producers’ ability to build and operate cleaner generators using wind and solar power. That comes at the cost of air pollution, including carbon emissions that contribute to global warming.
Needless to say, nuclear power plants have their own drawbacks. First, even the best ones produce radioactive waste that cannot be removed during a human lifetime. But nuclear power plants don’t burn carbon or spew tiny particles into the atmosphere that can damage your lungs. While building new nuclear power plants and gaining regulatory approval is difficult and expensive, older nuclear power plants are in high demand.
But for those deeply involved in environmental issues related to nuclear power, it’s daunting to think that the AI boom will make nuclear energy stocks the new rich.
insatiable needs
As a reporter for Newsday on Long Island, I covered nuclear power in the early 1980s. This was when nuclear power entered a period of rapid decline from which it would never fully recover.
Following the Three Mile Island disaster, many people around the world feared a nuclear holocaust. Bipartisan opposition led to the decommissioning of a completed but never fully operational commercial nuclear power plant in Shoreham, on Long Island’s picturesque north shore. The state purchased the Long Island Lighting Company, the blue-chip company that once built Shoreham.
At the time, I never dreamed that in 2024, artificial intelligence would spark a stock market boom for utilities with nuclear assets. It was beyond my ability to predict, and probably beyond any human ability to predict. AI may one day be able to help make such predictions, but I don’t think it’s ready for that right now. But AI’s thirst for energy is already insatiable.
Energy is required in two main ways. First, advanced AI runs primarily on supercomputers filled with Nvidia equipment. Supercomputers consume enormous amounts of power to “train” AI systems, processing data that tells them things, embellishes them, and transforms them.
Second, these supercomputers require enormous power so that AI chatbots can respond to questions and perform searches. Even a simple search in a bot like ChatGPT is energy-intensive, whether it’s asking a chatbot for Spanish homework, cooking recipes, or information about stocks.
How much power is it exactly? I spoke with Jesse Dodge, a senior researcher at the Allen AI Institute, a nonprofit in Seattle who has worked extensively on this topic. He and his colleagues estimated the kilowatt-hours consumed by advanced Nvidia hardware through a simple search and summarized it as follows: “We estimate that one query to ChatGPT uses the same amount of power as lighting one light bulb for about 20 minutes.”
Additionally, AI searches that occur automatically and unintentionally, such as when you perform a traditional search on Google and are also provided with an AI answer, require significant amounts of energy, he said.
AI does some things well, but many searches return boring or meaningless answers. “AI hallucinates and its answers are often unreliable,” Dr. Dodge said.
The widespread use of AI by consumers consumes energy and other valuable environmental resources, such as water to cool power plants and data centers, but this is not justified by people thinking about it. He said it would be difficult to do so. “This is a growing problem.”
This is a global problem, but it is also a local problem. Consumers value quick answers, and having AI supercomputers and their power sources closer to the metropolitan areas where most consumers live will reduce delays.
We are well beyond the experimental stage of artificial intelligence, where a small number of researchers occasionally perform energy-intensive searches. This field is growing rapidly with millions of people using AI search and new applications emerging all the time. Therefore, many researchers believe that the total amount of energy consumption from AI will reach a significant amount globally, somewhere around the energy consumption of countries like Sweden or Argentina today, and perhaps by 2030. It is estimated that it consumes more energy than India. No one knows what AI will ultimately become.
I can’t say there are limits. There is considerable skepticism in some parts of the investment community. There is no doubt that many new AI applications are exciting, but it is not clear how useful or profitable the technology will be. In June, Goldman Sachs released a report with the provocative title, “The AI Generation: Spending too much, earning too little?”
As an investor, I avoid betting on the future of specific innovations and instead put my money into broad, low-cost index funds that track the market as a whole. I don’t have the ability to judge the ultimate potential of AI any more than I have the ability to assess the long-term future of nuclear power in the market. Is the current stock price surge the beginning of a long-term revival? I don’t pretend to know.
However, enthusiasm for artificial intelligence is still growing. AI requires energy, even if it comes from Three Mile Island. A backdoor bet on AI through a nuclear operator is currently a strange and creative alternative to the Nvidia acquisition.
